Local AI hardware
DGX Spark and GB10 AI systems compared
NVIDIA DGX Spark and its GB10 partner systems put a Grace Arm CPU, Blackwell GPU and shared system memory in a compact desktop. Compare the exact configuration, support and software requirements before shopping.
Compare systems
| System | Unified memory | Storage | eBay median · 128GB |
|---|---|---|---|
| NVIDIA DGX Spark | 128 GB | 1 or 4 TB configurations | $8,500 |
| ASUS Ascent GX10 | 128 GB | 1, 2 or 4 TB configurations | $7,050 |
| MSI EdgeXpert | 64 or 128 GB | Check configuration | $13,189 |
| Dell Pro Max with GB10 | 128 GB | Check configuration | |
| HP ZGX Nano G1n AI Station | 128 GB | Check configuration | $11,951 |
| Lenovo ThinkStation PGX | 128 GB | 1 or 4 TB configurations | |
| Acer Veriton GN100 | 128 GB | Up to 4 TB |
GB10 or a discrete GPU?
GB10 systems trade the flexibility of a conventional desktop GPU build for a compact, shared-memory platform. A model’s actual memory use depends on its weights, quantization, context length and runtime. The 128 GB pool is not all available to a model, and capacity alone does not establish inference speed.
A discrete GPU has its own VRAM and a different software and upgrade path. Compare the complete system, measured workload performance and power needs, not just peak AI operations. See our local AI GPU guide and Mac versus GPU guide.
Before you buy
Confirm memory, SSD, included power adapter, warranty and return policy. DGX OS is Linux-based; x86-only software or containers may need an Arm-compatible alternative. NVIDIA’s porting guide explains the shared-memory architecture and software environment. Networking two systems requires compatible hardware and software and does not automatically make every application use both machines.